Why do LLMs only quote certain parts of my content?
- Dane Frederiksen
- 7 hours ago
- 3 min read
Because LLMs are lazy. They don't summarize an entire blog post or video — they scan for the "spiky" parts: a clean stat, a quotable line, a chart with a clear takeaway. Those are the fragments they extract and repeat in cited answers.
Quick Answer
LLMs are not summarizing your full content — that is too expensive for them to do at scale.
They scan for "spiky" content: a clear stat, a tightly-worded answer, a quotable segment.
Your job is to write content with extractable, citation-ready segments, not flowing prose.
Video gets cited the same way. The LLM looks for quotable fragments in your transcript, not the topic of the conversation.
Why Jessica Hennessey says LLMs are lazy
Jessica Hennessey, founder of Resonate Online and BetterSites, builds the kind of content systems that get cited by LLMs. Her framing is one of the cleanest mental models in GEO: LLMs are lazy. That is the whole game.
When a person prompts ChatGPT or Perplexity with a real question, the model is not reading your blog post end-to-end and synthesizing the main argument. It does not have the budget for that. What it does is scan for the smallest possible chunk that answers the question well — a stat in a callout box, a single sentence that defines the term, a number with a label on a chart.
"LLMs are lazy. So the more clear, pointed, nicely tied in a box kind of content that you give them, the more likely they will be to repeat that fact." — Jessica Hennessey
The same rule applies to video. The LLM is not watching your 15-minute interview and writing a synopsis. If it has access to the transcript, it is searching for what Jessica calls "spiky" segments — a clean quote, a specific number, a moment where someone makes a sharp claim. That is what gets pulled into a cited answer.
This changes what good B2B content looks like. The instinct is to write thoughtful, flowing arguments that build to a conclusion. The reality is that LLMs cite content that is structured for extraction: subheads with the question, a one-sentence direct answer, a stat, a quote. Pretty prose is not the path. Spiky, extractable content is.
Frequently Asked Questions
What does Jessica Hennessey mean when she says LLMs are lazy?
She means LLMs do not summarize your full content. They scan for the smallest extractable chunk that answers the prompt — a stat, a quote, a defined term — and repeat that fragment in cited answers. Building content for extraction beats building it for narrative flow.
How do I write content that LLMs will actually cite?
Structure it for extraction. Lead each section with the question as a subhead. Give a one-sentence direct answer right after. Pull out stats into callouts or quotable lines. Use clear, bounded segments — what Jessica calls "spiky" content — instead of flowing prose that builds to a conclusion.
How do LLMs decide what to cite from a video?
The same way they decide for written content. They scan the transcript for spiky, citation-ready segments — a sharp quote, a specific stat, a moment where someone makes a clean claim. They do not summarize the video's topic. They pull the fragment that answers the prompt.
Watch the full interview
Watch Jessica Hennessey and Dane Frederiksen go deeper on why CEOs need scripted answers, how to microdose video for AI search, and what spiky content looks like in practice: the full interview with Jessica Hennessey.
Comments